positioning in bluetooth and uwbnetworks
TRANSCRIPT
Positioning in Bluetooth and UWBNetworks
Rafael Saraiva Campos
Coordenação de Engenharia de Computação – CEFET/RJ
Petrópolis – RJ - Brasil
Abstract. This article brings a comprehensive tutorial on radio-frequency (RF)
positioning in Bluetooth and Ultra-Wideband (UWB) Wireless Personal Area Networks
(WPANs), carrying out an extensive literature review, providing the reader an overview
of Bluetooth and UWB RF positioning state-of-the-art. It also delves into some key
issues, such as multilateration (MLAT) based solutions using received signal strength
and time-of-arrival, multi-slope linear regression, RF fingerprinting and the use of
clustering techniques to improve its localization accuracy, the interrelation between
Bluetooth versions improvements and new positioning capabilities, as well as
localization prospects in future UWB networks.
1. Introduction
Bluetooth has been around for more than 20 years now (Stallings, 2005). Initially
devised to replace with wireless links the serial cables connecting peripherals to
computers or mobile phones, Bluetooth has evolved to support low-power short-range
communications between a variety of mobile devices, from laptops and smartphones to
earphones and body sensors. Its latest version(Bluetooth 4.0), with its focus on very low
power consumption(thereby greatly improving battery lifetime) and reduced connection
delay, made Bluetooth technology particularly well suited for indoor positioning in
WPANs.
UWB is an emerging technology that offers unique advantages, as a large
bandwidth and a very high time-domain resolution. The latter feature allows the
application of parameters such as the channel impulse response(CIR) to locate the
mobile station(MS). The use of CIR, at least theoretically, turns multi-path propagation
and non-line-of-sight(NLOS) conditions - both serious impairments to RF positioning in
other networks - into an additional information for accurate RF localization. UWB
transmissions, due to strict power limits (FCC, 2002), have limited range, being an
option for high accuracy WPAN positioning (Lau, 2011).
This article is organized as follows: first, the basics of the Bluetooth technology
are presented in Section 2; Section 3 brings a literature review of positioning solutions
proposed for Bluetooth networks; Section 4 introduces the fundamentals of UWB
technology. Finally, Section 5 presents a literature review of localization techniques in
UWB networks.
2. Bluetooth Networks
Bluetooth began as a low-power short-range wireless technology to replace
cables between desktop computers and its peripherals, such as keyboard and mouse.
Therefore, at its original conception back in 19941, it was designed to replace serial
RS-232 cables. Since then, it has greatly evolved, being incorporated into a vast number
of devices, such as laptops, smartphones, watches, among others. The Bluetooth
specifications are written and maintained by the Bluetooth Special Interest Group (SIG),
and include radio protocols and profiles2. Bluetooth is an open and royalty-free
standard, which is one of the reasons why it is the de facto standard for communication
in WPANs.
A Bluetooth WPAN is called a piconet, and comprises a master device and up to
seven active slave devices. A piconethas a star logical topology, so all transmission must
be relayed by the master - slave devices cannot communicate directly, except during the
discovery phase, when a Bluetooth device send inquiries to find nearby Bluetooth
devices. Bluetooth transmissions use the 2.402 - 2.480 GHz band, which is within the
Industrial, Scientific and Medical(ISM) 2.4 GHz unlicensed band. To reduce co-channel
interference and interference from external sources(such as WiFi, which also uses the
2.4 GHz ISM band), Bluetooth uses Frequency Hopping Spread Spectrum (FHSS) with
79 1-MHz channels and Time Division Multiplexing(TDM) with 625sec time slots.
The master defines the piconet synchronization clock (using its own internal clock) and
the frequency hopping sequence (FHS) that is going to be used by all devices in the
piconet, as well as the channel allocation for the slave devices. Transmissions hop from
one frequency to another every time slot. Frames might span up to 5 time slots. The
Bluetooth core version 1.0 uses Gaussian Frequency-Shift Keying Modulation (GFSK),
with data rates of up to 721 kbps.
2.1. Bluetooth Power Classes
Bluetooth devices can be classified into three power classes, as listed in Table1. The
power levels are specified at the antenna connector of the Bluetooth device. If the
device does not have a connector, a reference antenna with 0 dBi gain is assumed. Class
1 devices shall implement power control, with step sizes ranging from 2 to 8 dB, being
able to control their output power from 100 mW down to 2.5 mW. Between 2.5 mW and
1 mW, output power control is not used(Bluetooth Special Interest Group, 2015).
Table 1 - Bluetooth Power Classes
Power
Class
Max.
Output
Power
Min.
Output
Power
1 100 mW 1 mW
2 2.5 mW 0.25 mW
3 1 N/A
2.2. Bluetooth Protocol Architecture
The Bluetooth architecture has three key elements: the profiles, the host and the
controller. The profiles specify communication interfaces for Bluetooth compatible
1 JaapHaartsen, an Ericsson Engineer, in cooperation with Sven Mattison, designed the original Bluetooth
specification. This wireless technology was named after the tenth-century King Harold Blatand - i.e., Harald Blue
Tooth - who ruled over Denmark and Norway after uniting the Scandinavian tribes into a single kingdom, the same
way Bluetooth unites several protocols and devices into a WPAN. 2 A profile is a set of protocols components required for the correct set up of communicating applications.
devices. Some examples of Bluetooth profiles are the cordless telephone profile (CTP)
-which enables cellular phones to communicate with computers, acting as cordless
phones - headset, dial-up networking profile(emulates a modem over a cellular phone)
and file transfer profile (with the same functions of the file transfer protocol over
Internet protocol) (Labiod et al, 2007). The host is the hardware/software that supports
those profiles and that interfaces with the controller. The controller implements the Link
Manager Protocol (LMP) and comprises the Bluetooth baseband processor(which
provides services such as error correcting, hop sequence selection and security) and
radio(that carries out modulation, demodulation, power control, among other functions).
The host-controller interface(HCI) is the command interface to the Bluetooth controller
- also referred to as Bluetooth module - and allows access to hardware status and control
registers.
Figure 1 shows the Bluetooth protocol stack. TCS (Telephone Control Protocol
Specification) provides telephony services. SDP (Service Discovery Protocol) allows
Bluetooth devices to find out the services supported by other Bluetooth devices. OBEX
(Object Exchange) is a data communication protocol. RFCOMM emulates an RS-232
serial interface. L2CAP (Logical Link Control and Adaptation Protocol) multiplexes
data received from upper layers and adapts packet sizes between layers. The HCI allows
exchanges between the host and the Bluetooth module. The LMP controls and
configures links to other devices. The Baseband controls the physical link over the
Bluetooth radio.
Figure 1 - Bluetooth Protocol Stack
2.3. Bluetooth Evolution Path
Since 1999, when version 1.0 was issued, Bluetooth SIG has adopted several
enhancements to the core specification(which serves as a uniform structure for devices
interoperability), as listed in Table2.
Core version 1.2 introduced Adaptive Frequency Hopping(AFH), which aims at
reducing interference in Bluetooth transmissions due to the shared use of the unlicensed
2.4 GHz ISM band with other technologies, such as WiFi. The key idea is to identify
``bad'' channels and remove them from the hopping list. The channel assessment - i.e.,
the process of identifying the channels with high levels of interference(the ``bad''
channels) - is not defined in the SIG specification, so it is up to the Bluetooth device
manufactures to select which parameter(s) to use in the assessment. The most commons
are RSS and Packet Error Rate (PER): channels with high RSS levels and/or high PER
are removed from the hopping list. However, according the SIG specification, the FHS
must have at least 20 channels.
Table 2 - Milestones in the Bluetooth Evolution Path.
Core
Version
Issue
Year
Major Improvements
1.0 1999 -
1.2 2003 Adaptive frequency hopping,
Inquiry-based RSSI
2.0 2004 2.1 Mbps peak data rates
2.1 2007 3.0 Mbps peak data rates
3.0 2009 24 Mbps peak data rates
4.0 2010 Lower Energy Consumption,
Broadcasting,Lower connection latency
Prior to version 1.2, the obtaining of Received Signal Strength (RSS) in
Bluetooth had two relevant limitations (Wamg et al, 2013):
1) to obtain the RSS indicator (RSSI) of an anchor device, the target device had to
connect to it (connection-based RSSI). This introduced a delay that could render
triangulation-based or fingerprinting location methods unusable, particularly for
moving targets. In a 2003 study, the delay to get the RSS of a single node was 10.5
seconds - 5.3 seconds for the device discovery plus 5.2 seconds to connect to the
device (Hallberg et al, 2003). The total delay exceeded 15.4 seconds with only two
nodes. The longest delay to obtain a position fix reached 31.3 seconds using five
anchor nodes. After this interval, a person carrying a target Bluetooth device and
walking at a regular speed of 1.2 meters/second would be 38 meters away from the
location where the position request was initiated;
2) the connection-based RSSI was defined as follows: if the received power was
within the Golden Receive Power Range (GRPR) - a 20 dB wide interval that is
assumed as the ideal received power range - the RSSI value was set to 0 dB
(Bluetooth Special Interest Group, 1999). Within this interval, the RSSI was
constant;therefore, no relationship could be established between received power
and distance, for ranging and localization purposes. Out of the GRPR interval, the
RSSI was reported in dB above (positive values) the GRPR upper limit, or
below(negative values) the GRPR lower limit. Another problem was that the GRPR
lower limit was defined as a function of the receiver sensitivity3, so this parameter
was device specific (Bekkelien, 2012).
Core version 1.2 addressed the two aforementioned limitations, by introducing a
new HCI command called INQUIRY_WITH_RSSI, which allowed including the RSSI
in inquiry responses during the discovery phase(Bluetooth Special Interest Group, 2003).
Inquiry-based RSSI made it possible to obtain the RSSI without the need to connect to
another device, which reduced delay and overhead. Besides that, inquiry-based RSSI,
unlike connection-based RSSI, was not defined in relation to the GRPR lower and upper
limits, so the RSS could be reported in dBm, allowing the use of this parameter for
ranging and triangulation-based positioning.
3 In the Bluetooth core specification, the receiver sensitivity is defined as the received power level for which the Bit
Error Rate(BER) is equal to 0.001.
Core version 2.0 came along with Enhanced Data Rates(EDR), which achieved
peak data rates of 2.1 Mbps using Differential Quadrature Phase-Shift Keying(DQPSK)
modulation. This rate is almost three times higher than the previous version maximum
throughput of 721 kbps using GFSK. Core version 2.1 achieved up to 3 Mbps using
8-DPSK(Differential Phase-Shift Keying) (Labiod et al, 2007). Bluetooth controllers
compliant with core version 3.0+HS(High Speed) incorporate a WiFi device, reaching
up to 24 Mbps over WiFi links(Bluetooth Special Interest Group, 2009).
Core version 4.0 - also referred to as Bluetooth Low Energy(BLE) - brings
important improvements to Classical Bluetooth(the previous versions), mostly in energy
consumption optimization and faster connections. BLE implements features such as
smart host control and adjustable message length to reduce average, peak and idle mode
power consumption down to 20 times, in relation to Classical Bluetooth, allowing BLE
devices with small batteries to operate for a year or more (LitePoint, 2012). The smart
host control places much more intelligence on the Bluetooth controller, enabling the
host to sleep longer periods, being woken up by the controller only when some action
needs to be performed. This saves energy in comparison to Classical Bluetooth, where
the host usually consumes more power than the controller does. Adjustable message
length feature also saves energy, by encapsulating messages into longer packets - fewer
packets equals less overhead, reducing power consumption by the Bluetooth radio. BLE
achieves faster connections by using advertising mode, when the Bluetooth device sends
broadcast messages that speed up nearby devices discovery, pairing and connection
(Cinefra, 2013). BLE, unlike previous versions, is not backward compatible. For this
reason, the single mode devices - that support only BLE - cannot communicate with
Classical Bluetooth devices. Dual mode devices support both technologies, at the cost of
lower battery lifetime.
3. Positioning in Bluetooth Networks
The ubiquity of Bluetooth devices - laptops, smartphones, among others - coupled with
the low cost and long battery lifetime of Bluetooth modules(particularly with the advent
of BLE), makes Bluetooth networks a promising alternative for RF positioning,
especially in indoor environments. Similar to WiFi localization research, the focus in
Bluetooth positioning is placed in the evaluation of solutions that rely only on the
pre-existing network infrastructure, i.e., that do not require any additional hardware to
operate. This enables a rapid and cheap deployment of the location system, increasing
its acceptability4.
Initially, the Bluetooth signal parameters available for localization purposes are
the Link Quality Indicator (LQI)5
, the Transmitted Power Level (TPL)6
, the
connection-based RSSI and the inquiry-based RSSI. However, the best option,
presenting a more predictable relationship with TX-RX distance, is the inquiry-based
RSSI(available from core version 1.2 onwards), which is expressed in absolute values
(dBm instead of dB) and is not subject to variations due to power control(as the
4The acceptability can be defined as the location system's ability to be smoothly integrated with pre-existing network
infrastructure (Bandara et al, 2004). 5 The LQI is a 8-bit unsigned integer - defined as a function of the BER - that expresses the perceived link quality at
the receiver. The conversion from BER to LQI is device-specific. 6Power control might be used in Bluetooth connected state to reduce interference with other Bluetooth modules and
to improve battery lifetime. As it is used only in connected state, inquiry-based RSSI is not affected by it.
connection-based RSSI and TPL). LQI is related to the BER, but the conversion from
BER to LQI is device/manufacturer specific. Besides that, the LQI is subject to
variations due to channel conditions, such as interference, and as a result is not a viable
alternative for ranging and positioning (Hossain and Soh, 2007). Connection-based
RSSI is constant(0 dB) - and thereby independent of the TX-RX distance (Hallberg et al,
2003) - when the received power lies within the GRPR, which effectively also excludes
it as a feasible choice for ranging and localization. For example, assuming a path-loss
slope n=3(a typical value for 2.4 GHz RF propagation at indoor environments), a
Bluetooth device would report the same connection-based RSSI at 1 meter and 5 meters
from the Bluetooth beacon. However, Bandara developed a positioning system in
Bluetooth 1.1 using connection-based RSSI and installing variable attenuators at the
Bluetooth beacons(Bandara et al, 2004). The additional loss introduced by the
attenuators would then shift the received power to a level below the GRPR, where the
RSSI would be linearly related to the received power(and for that reason could be used
as a distance estimator for MLAT). A single beacon with variable attenuators and four
antennas - placed at the corners of a 4.5 x 5.5 m2 room - was deployed, and 92% of the
test samples achieved location errors of 2 meters or less. Nevertheless, the proposed
system requires the installation of additional hardware, which would be costly and
time-consuming in a large-scale location system - for example, in a whole building -
reducing the system's acceptability7.
Most features related to MS position location in WiFi networks are also
applicable to Bluetooth positioning: it is more relevant in indoor environments8;
round-trip-time(RTT) estimates are not available(which excludes time-of-arrival based
MLAT as a possible location technique); the Bluetooth beacon antennas are
omnidirectional, ruling out angle-of-arrival based triangulation, unless additional
hardware is deployed(i.e., directional antennas). The proposed solutions for Bluetooth
localization rely mostly on RSS-based MLAT(Wamg et al, 2013; Hallberg et al, 2003;
Cinefra, 2013; Fernandez et al, 2007; Chen et al, 2014), fingerprinting (Bandara et
al,2004; Zhang et al, 2013; Disha and Khilary, 2013), or on a combination of both
(Subhan et al, 2011; Subhan et al, 2013)9. In Bluetooth positioning, unlike in WiFi
networks, there are many MLAT based location solutions. This happens because, due to
the shorter range of Bluetooth beacons, this technique is being used mostly to locate
devices within each room. In that case, with the absence of obstacles such as walls
along the propagation path, it is easier to model the path-loss using empirical models,
allowing the use of RSS-based MLAT with good results. Proximity (i.e., returning as the
MS position estimate the coordinates of the nearest Bluetooth beacon) and centroid
methods are also used (Wamg et al, 2013).
3.1. RSS-based MLAT solutions
There is a vast amount of published papers on Bluetooth positioning using RSS-based
7Nevertheless, prior to core version 1.2, it would probably be the best option for Bluetooth positioning.
8In fact, it seems to be exclusively used for indoor localization, which is quite reasonable, if one considers the very
limited range of Bluetooth devices(with the exception of Class 1 devices), if compared to the range of WiFi access
points. 9Agrawalprovided a brief Bluetooth positioning taxonomy that includes some other less commonly used techniques -
such as Fuzzy Logic (Agrawal et al, 2014).
MLAT, such as (Wamg et al, 2013; Hallberg et al, 2003; Cinefra, 2013; Fernandez et al,
2007; Chen et al, 2014). In all of them, the following requisites are fulfilled:
1) the target MS is within range of at least three reference nodes;
2) there is a mathematical model describing path-loss as a function of distance;
3) there is an algorithm to solve the resulting overdetermined non-linear equation
system. This section is devoted to studying in detail two implementations with
special features: real-time calibration (Fernandez et al, 2007) and ranging with a
two-slope model(Chen et al, 2014).
3.1.1. Real-time calibration
Fernandez proposed a system using real-time calibration of the path-loss to distance
mapping(Fernandez et al, 2007). This scheme is able to perceive the effect of
environmental changes that affect RF propagation (such as humidity, temperature,
presence of people, change of furniture location, closing and opening of doors, among
others), therefore improving localization accuracy. Besides using the RSS between the
target and the reference nodes (Bluetooth beacons placed at known coordinates) to
achieve a positioning fix - as in any other MLAT system - the solution proposed by
Fernandez also employs the RSS between the anchor nodes to obtain calibration factors
- called translation (from path-loss to distance) factors(Fernandez, 2007). These factors
are then applied to the path-loss between the target and anchor nodes to yield distance
estimates that will be used in the MLAT position fix. Consider a set of N Bluetooth
beacons placed at known locations. This yields a pairwise distances matrix D =
[dij]NxN,where di,j is the Euclidean distance between anchor nodes i and j. Assuming
that all beacons have a known constant output power10
Pt, the path-loss between the
anchor nodes can be calculated, subtracting (in the logarithmic scale) the RSS values
from Pt. Then, a path-loss matrix S = [si,j]NxNis obtained, where si,j is the path-loss
between nodes i and j11
. A matrix of translation factors =[i,j], also referred to as
dynamic mapping matrix, is given by:
1ΦSD (1)
wherei,j is the factor that translates the path-loss si,j between nodes i and j into their
pairwise distance di,j. As is expressed as a function of the path-loss, it reflects the
variations in the propagation conditions, acting as a set of calibration factors in the
MLAT process. With calibration, the estimated distances between the target node and
the N Bluetooth beacons at any given moment are provided by vector
sd ˆ (2)
wheres is a N-element vector containing the RSS values from the N anchor nodes,
measured at the target device location. The following non-linear system is then
10
This is a reasonable assumption, as the proposed system uses inquire-based RSSI values, and replies to inquiries
during the discovery phase are not subjected to power control. 11
Note that both matrices D and S are symmetric and that the elements of their main diagonals are zero.
obtained:
222
2
1
2
1
2
1
ˆ
...
ˆ
NNN dyyxx
dyyxx
(3)
where (xi,yi) are the coordinates of the ith beacon and id is the estimated distance
between the target node and the ith beacon. Due to NLOS conditions and/or inherent
limitations in the RSS measurement and reporting, this quadratic equation system has
no closed-form solution. Fernandez (Fernandez et al, 2007) used a gradient descent
method (Allison, 2001) to find iteratively the position estimate y,x that minimizes the
following sum:
𝑥 , 𝑦 = 𝑎𝑟𝑔 𝑚𝑖𝑛 (𝑥 − 𝑥𝑖)2 + (𝑦 − 𝑦𝑖)
2 − 𝑑𝑖 2
𝑁𝑖=1 (4)
To evaluate the location system, the authors deployed a testbed in a 5 x 10m2
room, with four Bluetooth 2.0 beacons. Two test positions were selected: the first was at
the room center, and the second close to one of the beacons. Table 3summarizes results
at the two test locations, with and without calibration for two window lengths12
. When
no calibration is used, the target node estimated location is given by a weighted centroid
method. At the first test location, the use of calibration increased accuracy by 40% for
the smaller window length. At the second test location, it achieved an accuracy
improvement higher than 50%for both window lengths.
Table 3 - Positioning errors in centimeters (Fernandez et al, 2007), with
and without dynamic calibration.
Test Window
Lengt
(sec)
WithoutCali
bration
WithCali
br.
1 30 115.4 69.7
1 60 100.9 73.2
2 30 110.4 49.7
2 60 108.5 51.0
3.1.2. Ranging with a Two-Slope Model
Empirical propagation models can be used to express the received power as a function
of TX-RX distance, as the one defined by
0
100 log10)()(d
dndPdP (5)
where P(d) is the RSS (in dBm) at d meters from the transmitter, d0 is a reference
12 The window length or integration interval is the time along which RSS samples are collected and averaged.
distance(in meters), n is the path loss exponent or slope. Assuming d0=1 meter, one has
ddP 10log)( (6)
whereβ - also referred to as the offset - is the RSS at the reference distance and =-10n
is the path-loss slope in dB/decade. Calibration campaigns are required to fine-tune such
models before they can be applied to a specific environment. During these campaigns,
at each selected measurement point, the distance d and the average RSS value are
registered. Then, linear regression can be used to estimate parameters βand.However,
propagation characteristics might differ in regions close to and far from the transmitter.
In such cases, the use of a single offset-slope pair will result in path-loss prediction
errors that will escalate with distance. This will ultimately result in larger ranging errors
and consequently in lower positioning accuracy for MSs in the far region. To minimize
that, a breaking point might be identified, separating the near and far regions, and linear
regression might be applied to each region to estimate different model parameters. The
two-slope model is then given by
𝑃 𝑑 = 𝛽1 + 𝛾1 log10(𝑑) , 𝑑 < 𝑑𝑏𝑟𝑒𝑎𝑘
𝛽2 + 𝛾2 log10 𝑑
𝑑𝑏𝑟𝑒𝑎𝑘 , 𝑑 ≥ 𝑑𝑏𝑟𝑒𝑎𝑘
(7)
where (β1,1) and (β2,2) are the offset-slope pairs in the near and far regions,
respectively, and dbreak is the break point distance(in meters) in relation to the transmitter.
Multi-slope linear regression has been originally used for path-loss predictions in urban
microcells (Iskander and Yun, 2002), but its use can be extended to indoor environments.
It is applied by Chen(Chen et al, 2014), after the collected RSS values(during the
calibration or training phase) at each point are Gaussian-filtered and then averaged. The
break-point - also referred to as critical point - is empirically defined. Offset-slope pairs
are obtained for each Bluetooth beacon in the test bed. In the test phase, Chen used a
moving average filter to smooth abrupt variations in received RSS, before ranging(Chen
et al, 2014). The ranges obtained using the aforementioned two-slope model result in a
non-linear equation system. Taylor series first-order expansion is used to linearize the
system. In an experiment with four Bluetooth beacons, 80% of the test samples
achieved a position error lower than 1.5 meters.
By comparing Figures 2c and 2d, it is possible to confirm that the use of the
optimum two-slope model - illustrated in Figure 2b - results in a significant reduction of
the ranging error(diminishing its average by more than 60% in that particular example),
which has a direct impact on MLAT positioning accuracy. Though such improvement
might not be attainable in all scenarios, it proves that a two-slope model can provide a
better RSS to distance mapping in some environments.
Figure 2 - (a) Linear regression without a break-point; (b) Two-slope
linear regression, with a break-point at 14 meters from the transmitter; (c)
Ranging errors when using single slope regression; (d) Ranging errors
when using two-slope regression.
3.2. Fingerprinting-based solutions
Unlike in cellular and UWB networks, RF fingerprinting in Bluetooth(as in WiFi) is
restricted to using only RSS values. RF fingerprinting techniques for WiFi and Wireless
Sensor Networks (WSNs) indoor localization can be extended to Bluetooth
networks13
.Besides that, most papers on Bluetooth positioning use RSS based MLAT, as,
due to the limited range of Bluetooth beacons and devices, MLATis being used mostly
to locate devices within each room. For this reason, there are fewer papers on Bluetooth
fingerprinting, with no significant new features. For that reason, this sectioncovers in
detail two special features applied to RF fingerprinting: pattern matching using signal
strength difference (Hossain et al, 2013) and search space reduction using K-means
clustering.
3.2.1. Pattern Matching using Signal Strength Difference
Manufacturing variations among Bluetooth devices might affect the way they measure
RSS values. As a result, at a given location and time, different devices might measure
distinct RSS values for the same Bluetooth beacon. For this reason, if the radio map14is
built from field measurements using a certain Bluetooth device, and another Bluetooth
device is to be localized, then the fingerprinting location accuracy using this radio map
may deteriorate. This is called the cross-device effect (Chen et al, 2006). To mitigate it,
instead of employing absolute RSS values in the correlation function, one might use
relative RSS values, i.e., their ranking or their signal strength differences(SSDs).
While absolute RSS values of a set of Bluetooth beacons might be quite
13 In fact, with BLE and its emphasis on low energy consumption and enhanced battery lifetime, Bluetooth
positioning might be treated in a very similar way as in ZigBee WSNs. 14 The radio map, also referred to as the Correlation Database(CDB), is the set of geo-referenced RF fingerprints
collected(or generated via propagation simulations) during the off-line training phase of the fingerprinting algorithm.
different when measured by distinct Bluetooth devices, their ranking or their SSDs are
more likely to remain the same. The greater stability of those two techniques across
devices derives from the assumption that the relationship between the input signal
strength at the receiving antenna and the RSS reported by the Bluetooth device is a
monotonically increasing function (Krumm et al, 2003). For better understanding,
consider two Bluetooth devices, MS1 and MS2, both placed at the same location, and
two Bluetooth beacons, A and B. The signals from these beacons reach both MS devices
with strengths sA and sB, respectively. The RSS values informed by MS1 are RSS1A
and RSS1B. The RSS values informed by MS2 are RSS2A and RSS2B. If the
relationship between the input signal strength and the RSS is a monotonically increasing
function, then, if sA>sB, RSS1A> RSS1B and RSS2A>RSS2B. While the reported
values (RSS1A,RSS1B) might be different from (RSS2A,RSS2B) , the ranking of A and
B RSS values is the same on both MSs. Besides that, one can also assume that the SSD
of the signals from beacons A and B will be the same on both devices (Hossain et al,
2013), i.e., that (RSS1A-RSS1B)=(RSS2A-RSS2B).
In ranking correlation the Spearman Rank Correlation coefficient (Mathworks,
2010)is used to minimize the cross-device effect when comparing the target
fingerprint(Tfing) with the reference fingerprint (Rfings). When using SSD for that
same purpose, the Tfing becomes
𝑇 𝑘 = 𝑅𝑆𝑆1 − 𝑅𝑆𝑆𝑘
…𝑅𝑆𝑆𝑁 − 𝑅𝑆𝑆𝑘
(8)
wherei≠k(k {1,...,N})15
, N is the number of Bluetooth beacons and RSSi is the
received signal strength from the ith beacon. The Rfingsare also built using Equation (8).
As Example 1 shows, even though there are 2)!2(N
N!
possible pairs in a set of N
beacons, there are only (N-1) independent SSD values. That is why the fingerprints
dimension when using SSD is (N-1), if there are N beacons. The effect of this slightly
smaller RF fingerprint dimension, in comparison to when absolute RSS values are used,
becomes negligible for N>5(Kaemarungsiand Krishnamurthy, 2004).
Example 1:Given 𝑆 =[-40 -55 -80 -45]Tcontaining the absolute RSS values(in dBm) of
four Bluetooth beacons measured by a Bluetooth mobile device at a single location,
show that the resulting SSD RF fingerprints have only 3 independent SSD values.
Solution:Table4 shows vector 𝑆 and the resulting SSD RF fingerprints. Note that
vector𝑇𝑘 , k = {1,…,4} is obtained subtracting the ith beacon RSS value from the RSS
values of the remaining beacons. All fingerprints have only three non-null elements.
Besides that, it is also possible to observe that
𝑇 2 = 𝑇 1 + 𝑆 1 − 𝑆(2) 1 1 1 1 𝑇
𝑇 3 = 𝑇 1 + 𝑆 1 − 𝑆(3) 1 1 1 1 𝑇
𝑇 4 = 𝑇 1 + 𝑆 1 − 𝑆(4) 1 1 1 1 𝑇 (9)
15 Parameter k identifies the beacon which is fixed, and whose RSS is subtracted from the RSS of the remaining
beacons.
where S(k) is the kth element of 𝑆 . This means that all other fingerprints can be
obtained as a linear combination of the first one. Therefore, there are only 3 independent
SSD values. Generalizing, if there are N beacons, then only (N-1) independent SSD
values can be obtained.
Table 4 - Vector of absolute RSS values(in dBm) and the resulting RF
fingerprints using SSD (in dB).
𝑆 𝑇1 𝑇2
𝑇3 𝑇4
-40 0 15 40 5
-55 -15 0 25 -10
-80 -40 -25 0 -35
-45 -5 10 35 0
Hossain deployed a Bluetooth testbed in a laboratory spanning an area of 214 m2, with 4
fixed Bluetooth beacons and 337 training locations(Hossain et al, 2013). There, they
reported a 3.4-meter average location error when using K-Nearest Neighbors (KNN)
with absolute RSS values. The average error dropped to 2.6 meters when KNN was
used with SSD.
3.2.2. Search Space Reduction using K-means Clustering
Search space reduction aims at reducing the time to produce a position fix in RF
fingerprinting, by limiting the Rfings stored in the radio map - or Correlation
Database(CDB) - that will be compared with the Tfing. It is crucial when the radio map
is too large - which is the case for location systems in metropolitan areas (Campos and
Lovisolo, 2009) or even in large multi-floor buildings (Campos et al, 2014). In
MS-based Bluetooth indoor fingerprinting, search space reduction might also help
reduce energy consumption - both of the MSs and, most importantly, of the fixed
beacons - which is in keeping with one of the key features of BLE - i.e., enhancing
battery lifetime. Besides CDB filtering, another alternative for search space reduction is
K-means (Marsland, 2009), which is an unsupervised clustering technique16
. With
clustering, the positioning fix is carried out in two phases:
1) the Tfing is presented to the trained classifier, which identifies the cluster that the
Tfing belongs to;
2) theTfing is compared only to the Rfings belonging to the selected cluster. The
selected cluster - i.e., the reduced search space - is the one whose center is the
closest(in the n-dimensional RSS space)to the Tfing.
During the offline training phase, first the number of clusters Nc must be defined
and initial values must be assigned to the Nc clusters centers. Then, the following steps
are taken:
1) the distances between each Rfing and each cluster center are calculated;
2) each Rfing is associated with the cluster whose center is the closest one;
3) at the end of each epoch the clusters centers are re-calculated, using the arithmetic
16
Disha and Khilary reported a median positioning error below 2 meters in a Bluetooth indoor test bed using RF
fingerprinting with K-means(Disha and Khilary, 2013).
mean of the Rfings of each cluster17
;
4) the distances between each Rfing and the new clusters centers are calculated; if any
Rfing switches clusters, then steps 1 to 3 are repeated; otherwise, the process ends,
yielding the clusters centers vectors.
4. UWB NETWORKS
UWB transmissions occupy large bandwidths and have very strict output power limits.
As a result, they are best suited to high-speed short-range communications. These
features, coupled with their very high time-domain resolution, make UWB a promising
alternative for high accuracy indoor WPAN positioning. UWB signals might be
generated using very short time-domain pulses(from 10 to 1000picoseconds), that
occupy large bandwidths(from hundreds MHz to several GHz) in the frequency domain.
This is called single-band UWB, also referred to as impulse radio UWB(IR-UWB).
Another alternative to generate UWB signals is to divide the available UWB band into a
number of non-overlapping channels18
. Transmissions are made simultaneously over
multiple carriers into those non-overlapping channels. This is called multi-band UWB,
also referred to as multi-band orthogonal frequency division multiplexing(MB-OFDM)
UWB (Lau, 2011). Each symbolis transmitted over several pulses(in IR-UWB) or over
several non-overlapping channels(in MB-OFDM UWB). UWB transmissions have a
low power spectral density and are spread over wide bandwidths, and do overlap with
frequency bands occupied by other services. This latter characteristic arose the concerns
of several segments of the telecommunications industry, afraid that the cumulative
effect of UWB signals(aggregated power) could cause harmful interference to
pre-existing systems, by raising the background noise19
. As a result, regulatory bodies,
such as the Federal Communications Commission(FCC) in the USA, issued
specifications for UWB systems operation.
4.1. Definition of UWB Technology
The classification of a communications system into narrowband, wideband or
ultra-wideband, shown in Table5 (Lau, 2011)is based on the fractional bandwidth Bf,
which is given by
𝐵𝑓 = 2 𝑓𝐻−𝑓𝐿
𝑓𝐻+𝑓𝐿 (10)
wherefH and fL are the upper and lower 10 dB points, respectively(i.e., those are the
frequencies at which the output power is 10% of the maximum output power, which
occurs at frequency fM, where fL<fM<fH). Any system with Bf 20% or that occupies a
bandwidth larger than 500 MHz (the bandwidth is given by (fH-fL)) is classified as a
17 At this point, another unsupervised clustering method called K-medians, instead of the arithmetic mean, uses the
median. The advantage of using the median is that it is a more robust estimator, less susceptible to outliers, i.e., input
vectors with much noise (Marsland, 2009). All other steps are equal for both K-medians and K-means. 18
MB-OFDM splits the UWB spectrum into 528-MHz wide bands. Each 528 MHz band comprises 128 carriers
modulated using Quadrature Phase-Shift Keying(QPSK) on Orthogonal Frequency Division Multiplexing(OFDM)
tones (Matin, 2010). 19 UWB signals are noise-like, due to their very large bandwidth and their very low power spectral density. For that
reason, depending on the attitude of regulatory bodies, license-free operation of UWB systems could be allowed
(Oppermann, 2006).
UWB system (Oppermann, 2006).
Table 5 - System type classification as a function of Bf
System
Type
Fractional
Bandwidth
Narrowband Bf< 1%
Wideband 1% Bf< 20%
Ultra-wideband Bf 20%
4.2. Overview of FCC UWB Regulations
The FCC issued the UWB First Report and Order in February 2002(FCC, 2002). It
authorized three classes of UWB systems:
1) imaging systems(such as through the wall imaging, ground penetrating radar and
medical imaging systems);
2) vehicular radar systems;
3) communication and measurement systems(among which RF positioning is
included).
These regulationsdefine, among other parameters, the frequency bands and the
emission limits for UWB systems. Initially, the band allocated for UWB is 7.5-GHz
wide, between 3.1-10.6 GHz (Oppermann, 2006). At that band, the maximum output
power for indoor applications is 0.5 mW, which corresponds to a power spectral density
of approximately -41.3 mW/MHz. For outdoor applications in that band, such as
vehicular radar, the maximum power spectral density is 10 dB lower. Table6summarizes
the emission limits(expressed as power spectral densities in dBm/MHz) for the
authorized UWB systems in different frequency bands (Breed, 2005).
Table 6 - FCC UWB emission limits (dBm/MHz)
Application
Frequency Band (GHz) 0.96-1.61 1.61-1.99 1.99-3.1 3.1-10.6 > 10.6
Ground Penetrating Radar -65.3 -53.3 -51.3 -41.3 -51.3
Surveillance Systems -53.3 -51.3 -41.3 -41.3 -51.3 Medical Imaging Systems -65.3 -53.3 -51.3 -41.3 -51.3 Indoor Comm. Systems -75.3 -53.3 -51.3 -41.3 -51.3 Vehicular Radar Systems -75.3 -61.3 -41.3 -51.3 -61.3
4.3. IEEE Task and Study Groups related to UWB
The Institute of Electrical and Electronics Engineers(IEEE) created in May 1999 the
IEEE 802.15 WPAN Working Group, which aims to provide standards for
low-complexity and low-power consumption wireless connectivity. Within this work
group, the following currently active(as of January 2015) task and study groups are
related to UWB:
IEEE 802.15 TG3d(Task Group 3d 100 Gbit/s Wireless): its objective is to
increase the data transfer speeds of 802.15.3 standard for imaging and
multimedia applications, reaching data rates of up to 100 Gbps;
IEEE 802.15 TG4r(Task Group 4r Distance Measurement Techniques): its
objective is to provide communications and high precision ranging/location
capability;
IEEE 802.15 SG3c(Study Group 3e Close Proximity High-Data Rate systems):
its objective is to explore the use of the 60 GHz band for WPANs;
Besides IEEE 802.15, another working group of interest to UWB is IEEE 802.19
Wireless Coexistence Working Group, which develops and maintains policies
addressing issues of coexistence with current standards and/or standards under
development. Due to the very large bandwidth of UWB, and as UWB systems use
frequency bands overlapping with other systems, several coexistence scenarios for
UWB devices have been submitted to IEEE 802.19 (Politano, 2006). It is also worth
mentioning that IEEE issued a specification, IEEE 802.15.4a, which expands the
original IEEE 802.15.4(WPANs) to encompass UWB technologies as well. However, it
has not been ratified yet.
4.4. UWB versus Bluetooth
UWB and Bluetooth can be seen as competing technologies for WPANs: both aim at
low power consumption, enhanced battery lifetime and data communications over short
distances. Bluetooth is already a well-established technology, with a well-defined
protocol stack. UWB, on the other hand, is still an evolving technology, and its
regulation is still not completely agreed upon. Besides that, several Medium Access
Control(MAC) protocols have been proposed for UWB20
, and there is still ongoing
research on this topic. In fact, there are many open issues in UWB MAC especially in
four areas: overhead reduction, multiple access, resource allocation, and quality of
service(QoS) provisioning(Zin and Hope, 2010). However, on the long run, UWB is
most likely to overcome Bluetooth for short-range communication applications, due to
the much higher data rates it can achieve. For example, in UWB WPANs, data rates of
up to 480 Mbps will allow high speed data transfer of multimedia content, an
application not supported by BLE (Matin, 2010).
5. POSITIONING IN UWB NETWORKS
As in narrowband and wideband systems previously studied, localization in UWB
networks can use geometric(i.e., triangulation techniques - circular and hyperbolic
MLAT;multi-angulation) and RF fingerprinting. However, not all those methods can
take full advantage of the UWB signals very high time-domain resolution.
5.1. Geometric positioning
5.1.1. Multi-angulation
Angle-of-Arrival(AOA) positioning, also referred to as Direction of Arrival(DOA),
requires the deployment of antenna arrays at the reference nodes, what increases the
localization system infra-structure cost. Besides that - and maybe more importantly -
AOA positioning accuracy is severely affected by multi-path and NLOS propagation
conditions. As a result, AOA is not a viable alternative for practical deployments of
UWB indoor localization systems (Yang, 2011).
5.1.2. TOA-based MLAT
Time-of-Arrival(TOA)-based MLAT, as in narrowband and wideband systems, is
negatively affected by multi-path and NLOS conditions. However, high time-domain
20
Zin and Hope provided a summary of proposed UWB MAC protocols (Zin and Hope, 2010).
resolution of UWB signals enable an accurate TOA estimation21
. Aside from detecting
the TOA as exactly as possible, it is also important to identify the channel conditions to
improve circular MLAT accuracy. Three basic channel conditions have been
categorized(Pahlavan et al, 1998):
1) Dominant Direct Path(DDP), when the line-of-sight(LOS) component(the first
received multi-path component) is the strongest;
2) Non-Dominant Direct Path(NDDP), when there is a LOS component, but it is
not the strongest;
3) Undetected Direct Path(UDP), when there is no LOS component, i.e., the direct
path between the transmitter and the receiver is obstructed.
First, the channel condition must be identified, primarily to find out if there is a
direct path(LOS). This first step is called NLOS detection. Second, if NLOS conditions
have been verified in the first step, NLOS mitigation techniques should be
applied.Schroederpresents several methods for NLOS detection(Schroeder, 2007). Lie
and See analyze some alternatives for NLOS mitigation(Lie and See, 2011). All these
measures help improving ranging(obtained from the TOA measurements), and, as a
result, the positioning accuracy as well. The high time-domain resolution of UWB
signals enhances the NLOS detection and mitigation capabilities of those algorithms.
5.1.3. RSS-based MLAT
RSS-based MLAT, even though a viable alternative for UWB localization, does not
benefit from the UWB high time-domain resolution, resulting in positioning accuracies
similar to those achieved by narrowband or wideband systems. For example, Amiot and
Laaraiedhdeployed four UWB receivers at fixed locations in a 350 m2 floor in an office
building(Amiot and Laaraiedh, 2013). A transmitter was moved along 302 positions
selected within that floor, generating IR-UWB signals in the 3-7 GHz band, with an
output power of 26 dBm. The RSS at each receiver, for each transmitting position, was
registered, together with the TX-RX distance. Those values were used to calculate
log-normal shadowing path-loss models for each receiver location, employing linear
regression. The final position estimate was obtained in a two-step process: first, the
room where the MS was located was identified using RSS-based fingerprinting; second,
the MS location within the selected room was estimated using RSS-based MLAT. In the
first step, a 98.7% room identification accuracy was reported. In the second step, a
median positioning error of approximately 3 meters was achieved.
5.2. RF Fingerprinting
CIR-fingerprinting takes full advantage of the huge UWB bandwidth and yields high
position accuracy. This method takes factors which impair positioning accuracy in
narrowband and wideband systems - multi-path propagation and NLOS conditions - and
turns it into an advantage to RF localization22
.
The CIR to the impulse transmitted by the ithanchor node, measured at the jth
measurement point (i.e., at the jth reference point of the radio map), is given by
21Yang analyzes several algorithms for TOA estimation in UWB networks (Yang, 2011). 22Some results show that CIR-fingerprinting location accuracy is higher in NLOS than in LOS conditions(Wasim and
Ben, 2007).
𝒉𝑖 ,𝑗 𝑡 = 𝑎𝑖 ,𝑗 ,𝑘𝑀𝑘=1 𝛿(𝑡 − 𝜏𝑖 ,𝑗 ,𝑘)𝑒−𝑗𝜃𝑖 ,𝑗 ,𝑘 (11)
where ai,j,k, i,j,k and i,j,k are the amplitude, phase and delay of the kth received
multi-path component from the ith anchor node transmission, measured at the jth
reference point23
; M is the maximum number of resolved multi-path componentsand (t)
denotes the impulse function. If at reference point j, less than M multi-path components
are detected from the impulse transmitted by anchor node i, then zero-padding is used to
complete vector hi,j, which will then have M elements, for any i=1,…,N and j=1,…,NR.
Note that N is the number of anchor nodes and NR is the number of entries in the radio
map.
Equation (11) shows that, in multi-path propagation conditions, the receiver
detects multiple delayed and attenuated copies of the transmitted impulse. By taking the
CIR with respect to different anchor nodes at the same position j, it is possible to build
an RF fingerprint given by
𝐑𝑗 = 𝐡1,j 𝑡 … 𝐡N,j 𝑡 (12)
Each element of the fingerprint is a M-element vector, so the CIR-based
fingerprint is a M x N matrix. The CIR is expected to carry the multi-path information
that is unique to each location.
Assume that the target device measures the CIR to the N anchor nodes, yielding
the target RF fingerprint given by
𝐓 = 𝐡1 𝑡 … 𝐡j 𝑡 (13)
wherehi is the CIR with respect to the ith anchor node. The target node position estimate
is provided by the coordinates of reference pointj, whose CIRs(stored in the reference
fingerprint Rj) maximize the sum of the CIR cross-correlation coefficients, i.e
𝑗 = 𝑎𝑟𝑔𝑚𝑎𝑥 𝐶𝑖,𝑗 𝑁𝑖=1 (14)
wherejW, W={1,…,NR}24
, NR is the number of georeferenced points in the radio map
and Ci,j is the CIR cross-correlation coefficient between the target(T(i) = hi) and
reference (Rj(i) = hi,j) CIRs with respect to the ith anchor node, given by
𝐶𝑖 ,𝑗 =𝐸 𝑚𝑛∗ −𝐸 𝑚 𝐸{𝑛∗}
𝐸 𝑚 2 − 𝐸{𝑚} 2 𝐸 𝑛 2 − 𝐸{𝑛} 2 (15)
wherem=hi,j and n=hi; n* denotes the conjugate complex of n and E{.} denotes the
expectation operator. To improve positioning accuracy, a form of KNN is used: instead
of returning only the coordinates of the reference point identified by index j , that
maximizes Equation (14), all reference points where the sum of the CIR
23 Each reference point, stored in the radio map or CDB, is georeferenced: there is a pair of coordinates (for planar
positioning) associated with it. 24 Note that W is simply the set of reference point indexes in the radio map.
cross-correlation coefficients is greater than a threshold Cth are selected25
. The target
node position estimate is then given by the arithmetic mean of the coordinates of the
selected reference points, or the position fix might return an ambiguity region, formed
by the coordinates of the selected reference points.
Wasim and Bencarried out indoor measurements in the FCC allocated UWB
band(3.1-10.6 GHz)(Wasim and Ben, 2007). Some locations have been selected for the
UWB impulse radio. Then, a set of measurement points was identified in the indoor
testbed. For each transmitter location, at each measurement position, 1601 discrete
frequency samples were taken over the 7.5-GHz wide band, and the Channel Transfer
Function(CTF) was calculated(Sindi, 2013). The CIR was obtained calculating the
Inverse Discrete Fourier Transform(IDFT) of the CTF. On average, the ambiguity
regions returned by the position fixes had radiuses of approximately 2 cm. Nevertheless,
such a high accuracy does not come without a price, as the training phase of
CIR-fingerprinting is time consuming, complex(as it requires specific equipment, such
as vector network analyzers) and the resulting radio map might be very large. For these
reasons - that are of paramount importance, if one considers a practical implementation
of such a positioning system - this method is more suited only to small-size indoor
environments.
6. SUMMARY
This paper presented the fundamentals of Bluetooth and UWB positioning. A brief
review of the Bluetooth core versions, along its evolution path, has been provided,
highlighting the improvements in the Bluetooth specification that were relevant to RF
positioning, such as inquiry-based RSSI. Two techniques to enhance RSS-based MLAT
in Bluetooth networks have been studied: real-time calibration and ranging with
two-slope models. Following, the basics of the UWB technology have been introduced:
the types of UWB signals(IR-UWB and MB-OFDM UWB), the allocated frequency
bands, the output power limitations, among other relevant information. Finally, the
applicability in UWB networks of geometric and scene analysis localization techniques
have been analyzed, and CIR-fingerprinting, which achieves average location errors
around 2 centimeters, has been presented.
REFERENCES
AGRAWAL, S. et al. 2014. Indoor Localization based on Bluetooth Technology: A
Brief Review. International Journal of Computer Applications, vol. 97, number 8,
31-33.
AMIOT, N. and LAARAIEDH, M. 2013.Refined characterization of RSSI with
practical implications for indoor positioning. 10th Workshop on Positioning, Navigation
and Communication (WPNC).
BANDARA, U. et al. 2004.Design and implementation of a Bluetooth signal strength
based location sensing system. IEEE Radio and Wireless Conference.
BEKKELIEN, A. 2012. Bluetooth Indoor Positioning. Master’s Thesis. University of
25 As |Ci,j| 1, for any i=1,…,N and j=1,…,NR, with N anchor nodes, we would have 0 Ci,j< N. A typical condition
would be Cth = 0.9N. However, the value of Cthmust be carefully chosen, according to the specific propagation
conditions at the place where the positioning system is to be deployed. Its value could then be determined using the
fingerprints collected during the training phase.
Geneva.
BLUETOOTH SPECIAL INTEREST GROUP. 1999.Bluetooth Spec. (1.0B), 693.
BLUETOOTH SPECIAL INTEREST GROUP. 2003.Bluetooth Spec. (1.2), v.2, 573.
BLUETOOTH SPECIAL INTEREST GROUP. 2009.Bluetooth Spec. (3.0+HS), v.5, 14.
BLUETOOTH SPECIAL INTEREST GROUP. 2015.About Bluetooth SIG [Online].
Available: bluetooth.com/ Pages/about-bluetooth-sig.aspx.
BREED, G. 2005. A Summary of FCC Rules for Ultra Wideband Communications.High
Frequency Electronics, 42-44.
CAMPOS, R. S. and LOVISOLO, L. 2009. A Fast Database Correlation Algorithm for
Localization of Wireless Network Mobile Nodes using Coverage Prediction and Round
Trip Delay. 69th IEEE Vehicular Technology Conference.
CAMPOS, R. S. et al. 2014. Wi-Fi Multi-floor Indoor Positioning considering
Architectural Aspects and Controlled Computational Complexity. Expert Systems with
Applications, vol. 41, number 14, 6211-6223.
CHEN, M. et al. 2006. Practical Metropolitan-Scale Positioning for GSM Phones. 8th
International Conference on Ubiquitous Computing.
CHEN, Z. et al.2014. RSSI Based Bluetooth Low Energy Indoor Positioning.
International Conference on Indoor Positioning and Indoor Navigation.
CINEFRA, N. 2013. An adaptive indoor positioning system based on Bluetooth Low
Energy RSSI. Master’s Thesis. Politecnico de Milano.
DISHA, A. and KHILARI, G. 2013.Fingerprinting Based Indoor Positioning System
using RSSI Bluetooth. International Journal for Scientific Research and Development,
vol. 1, n. 4, 889-894.
FCC, First Report and Order FCC 02-48. 2002 [Online]. Available:
transition.fcc.gov/Bureaus/Engineering\_Technology/Orders/2002/fcc02048.pdf
FERNANDEZ, T. M. et al. 2007. Bluetooth Sensor Network Positioning System with
Dynamic Calibration.4th International Symposium on Wireless Communication Systems
(ISWCS).
HALLBERG. J et al. 2003. Positioning with Bluetooth. 10th International Conference
on Telecommunications (ICT).
HOSSAIN, A. and SOH, W.-S. 2007.A Comprehensive Study of Bluetooth Signal
Parameters for Localization. 18th International Symposium on Personal, Indoor and
Mobile Radio Communications (PIMR).
HOSSAIN, A. et al. 2013. SSD: A Robust RF Location Fingerprint Addressing Mobile
Devices' Heterogeneity. IEEE Transactions on Mobile Computing, v.12, 1, 65-77.
ISKANDER, M. F. and YUN, Z. 2002. Propagation Prediction Models for Wireless
Communication Systems. IEEE Transactions on Microwave Theory and Techniques, vol.
50, number 3, 662-673.
KAEMARUNGSI, K. and KRISHNAMURTHY, P. 2004. Modeling of indoor
positioning systems based on location fingerprinting. 23rd
Annual Joint Conference of
the IEEE Computer and Communications Societies (INFOCOM 2004).
KRUMM, J. et al. 2003. RightSPOT: A Novel Sense of Location for a Smart Personal
Object. Proceedings of Ubicomp 2003.
LABIOD, H. et al. 2007. Wi-Fi, Bluetooth, ZigBee and WiMax, Springer, 1sted.
LAU, H. K. 2011. High-Speed Wireless Personal Area Networks: An Application of
UWB Technologies. In Novel Applications of the UWB Technologies, 1st ed., Boris
Lembrikov (Editor), InTech, ch. 7.
LIE, J. P. and SEE, C.-M. 2011. NLOS Mitigation Techniquesfor Geolocation.
InHandbook of Position Location: Theory, Practice,and Advances, 1sted., S. A. Zekavat
and R. M. Buehrer (Editors), John Wiley and Sons, ch. 17.
LITEPOINT. 2012. Bluetooth Low Energy (Tech. Report).
MARSLAND, S. 2009. Machine learning: an algorithmic perspective, CRC Press.
MATIN, Mohammad. 2010. Ultra Wideband Preliminaries. In Ultra Wideband, 1st Ed,
Boris Lembrikov (Editor), Intech, ch.1.
OPPERMANN, I. 2006. Introduction. In UWB Communication Systems: A
Comprehensive Overview, 1st Ed, Maria-Gabriella Di Benedetto et al (Editors), Hindawi
Publishing Corporation, ch.1.
PAHLAVAN, K. et al. 1998. Wideband radio propagation modeling for indoor
geolocation applications. IEEE Communications Magazine, vol. 36, number 4, 60-65.
SCHROEDER, J. et al. 2007. NLOS detection algorithms for Ultra-Wideband
localization. 4th Workshop on Positioning, Navigation and Communication (WPNC).
MATHWORKS. 2010. Pairwise distance between pairs of objects [Online].
Available:.mathworks.de/access/helpdesk/help/toolbox/stats/pdist.html.
STALLINGS, W. 2005. Wireless Communications and Networks, Pearson P. Hall, 2nd
ed.
SUBHAN, F. et al. 2011. Indoor positioning in Bluetooth networks using fingerprinting
and lateration approach. International Conf. on Information Science and Applications.
SUBHAN, F. et al. 2013. Kalman Filter-Based Hybrid Indoor Position Estimation
Technique in Bluetooth Networks. International J. of Navigation and Observation.
WAMG, Y. et al. 2013. Bluetooth positioning using RSSI and triangulation methods.
IEEE Consumer Communications and Networking Conference (CCNC).
WASIM, M and BEN, A.2007. Wireless Sensor Positioning with Ultrawideband
Fingerprinting. The Second European Conference on Antennas and Propagation.
YANG, L. and XU, H. 2011. Wireless Localization using Ultra-Wideband Signals. In
Handbook of Position Location: Theory, Practice, and Advances, 1st Ed, S. A. Zekavat
and R. M. Buehrer (Editors), John Wiley and Sons, ch.8.
ZHANG, L. et al. 2013. A Comprehensive Study of Bluetooth Fingerprinting-Based
Algorithms for Localization. 27th International Conference on Advanced Information
Networking and Applications Workshops (WAINA).
ZIN, M. and HOPE, M. 2010. A Review of UWB MAC Protocols. Sixth Advanced
International Conference on Telecommunications (AICT).